assaybench: resources_servers: assaybench: entrypoint: app.py domain: knowledge description: AssayBench -- rank the hit genes of a CRISPR screen from its plain-text description (arXiv:2605.10876) value: Tests whether a model can link mechanistic biology to measured phenotypic outcomes, scored with the paper's adjusted nDCG verified: false # RankingMetrics settings from the authors' `assaybench` package, as used for the paper. k_values: [5, 10, 20, 50, 100] reward_metric: adjusted_ndcg@100 use_gene_mapper: true # Upstream's quality gate (re-scan the raw reply below this many genes); 0 = off, as upstream. min_expected_genes: 0 strip_think_blocks: true assaybench_simple_agent: responses_api_agents: simple_agent: entrypoint: app.py resources_server: type: resources_servers name: assaybench model_server: type: responses_api_models name: policy_model datasets: - name: example type: example # Pre-materialized with benchmarks/prompts/eval/assaybench/paper.yaml; do not pass +prompt_config. jsonl_fpath: resources_servers/assaybench/data/example.jsonl license: MIT assaybench_environment_server: environment_servers: single_agent_turn_legacy: entrypoint: app.py resources_server: type: resources_servers name: assaybench agent_server: type: responses_api_agents name: assaybench_simple_agent resources_tool_transports: - direct_http